去年12月，OpenAI在其模型规范中新增了针对未成年人的基本原则。此前数月，该公司发布了开源推理模型gpt-oss-safeguard，旨在协助开发者实施安全条款并对内容进行安全分级。与传统安全分类流程不同，该模型可直接接入平台安全策略，在判断输出内容适宜性的同时理解政策意图。
Researchers have highlighted how weaknesses in agent-driven environments can allow malicious instructions, prompt injection attacks or poisoned data to propagate rapidly across interconnected systems. In enterprises where agents are connected to sensitive data, financial systems or operational infrastructure, even small governance gaps can escalate into material risk.
，更多细节参见向日葵下载
Поделитесь мнением! Поставьте оценку!，推荐阅读Line下载获取更多信息
Jimmy Lai, 78, British citizen and Hong Kong media tycoon
图片来源：Toru Hanai / Reuters